Frequently asked questions
AI Skills: Configuration & Troubleshooting
This category covers issues related to the setup, performance, and errors within specific AI skills like Document Information Extraction (DIE), Document Splitting, and Classification.
Why are my AI Skill's labels not loading, or why am I getting an "Unexpected server error" when trying to edit the skill?
This is frequently caused by using forbidden characters (e.g., ., [ ], , ;, "") in label names. These characters have a special meaning for the system's internal processing. A platform fix has been released that allows you to access skills with these legacy naming issues so you can edit the labels to remove the forbidden characters. For all new skills, avoid using these characters in label names to prevent downstream problems in areas like Ground Truth and API outputs.
My Document Splitting skill is failing. What are the most common causes?
Failures in Document Splitting often stem from issues with the input files. Common causes include:
- Corrupted Files: The file, especially if it's a PNG or a PDF converted from another format, may be corrupted. Please verify the file's integrity.
- Mismatched File Type: The file's content might not match its extension (e.g., a file saved with a .png extension that is not a valid PNG image). This can trigger a File content type mismatch error. You can resolve this by using a Code node in your workflow to correct the file's MIME type before it reaches the Document Splitting node.
- Empty Documents: Processing completely empty or unfilled forms can cause an internal error in the AI model. This is a known issue being addressed by the R&D team.
Why are the confidence scores from my extraction skill inconsistent or very low?
Confidence scores can fluctuate based on model versions and document quality. To help the support team investigate, please provide the direct URL to your AI skill and clarify if the documents you are testing with have been added to your Ground Truth. Consistent use of Ground Truth helps stabilize and improve model performance over time.
How do I create nested or dynamic table structures in my extraction skill?
Currently, the platform does not support dynamically creating nested label groups (i.e., a group within a group). All labels must be defined manually. As a workaround, you can simulate a nested structure by adding a "parent" column to your table. For example, instead of nesting items under 'Person 1', 'Person 2', etc., you can add a column named 'Person Index' and populate it with '1' or '2'. This allows you to reconstruct the desired hierarchy from the flat JSON output.
I'm getting an error when using the "Try Skill" feature.
The "Try Skill" feature may not be available for all skill types or could experience temporary issues. For example, the Document Splitting skill's "Try Skill" view is currently under development. If you encounter an "internal error," it could be due to an issue with the underlying model (e.g., processing an empty document). In such cases, please report the issue with the skill URL.
Workflow & Agentic Flow Builder
This section addresses common problems encountered while building and running automation workflows, including node configuration, execution control, and webhooks.
How can I handle errors globally in a workflow, similar to an "Error Trigger" in Agentic Flow Builder?
While a dedicated global "Error Trigger" node is not available, you can configure error handling on a per-node basis using the "Settings" tab within each node and defining what should happen if an error occurs. For more complex, workflow-wide error management, this feature is being considered for future updates.
An old workflow keeps running even though I deactivated it. How do I stop it?
This issue is often caused by a deprecated "Interval" node. This node has been replaced by the more reliable "Schedule" node. To fix this, please review your old workflows, remove any instances of the "Interval" node, and replace them with the "Schedule" node.
I'm unable to find or select my Agentic Skills in a workflow node.
This was a temporary platform bug that has since been resolved. If you encounter this, please try a hard refresh of your browser (Ctrl+F5 or Cmd+Shift+R) or log out and back in
I need to use advanced scripting with modules like crypto for JWT signing. How can I do this in a Code node?
The Function/Code node runs in a sandboxed environment with limited access to built-in Node.js modules for security reasons. The recommended workaround is to host a small, external microservice or serverless function that performs the cryptographic operation and then call it from your DeepOpinion workflow using the "HTTP Request" node.
My webhook isn't receiving URL parameters.
The platform's webhook trigger fully supports URL query parameters. Please ensure your URL is formatted correctly (e.g., .../webhook/your-id?param1=value1¶m2=value2). The received parameters will be available in the output of the webhook trigger node under the query object.
Ground Truth & Model Training
This category is for issues related to uploading data, annotating documents, and optimizing models using the Ground Truth interface.
My documents are stuck "syncing" or fail to upload to Ground Truth
This can happen for two main reasons:
- Platform Issues: There may be temporary system delays. You can try the upload again after a short period.
- Label Naming Issues: The most common cause is the use of forbidden characters (e.g., . [ ] ;) in the skill's label names. These characters can corrupt the data structure, leading to failed syncs. Please review and correct your label names to resolve this.
The Ground Truth editor is glitchy. Values are flashing or my edits are being deleted.
These UI glitches are also symptoms of using forbidden characters in label names. When the front-end tries to process a label like Details.Name, it can lead to instability. The solution is to correct the label names by removing the special characters. Another reported issue is that adding a new row to a table sometimes deletes an existing row; this has been escalated to the engineering team.
Platform, UI & Account Management
This section covers general platform issues, such as dashboard visibility, login problems, and permission errors.
Why are the charts on my dashboard, like "Credit Consumption," not loading?
This issue is typically encountered by users on corporate networks or those using SSO. It happens when the organization's firewall blocks the URLs that serve the chart data. Please contact your IT department to whitelist all DeepOpinion domains. In some cases, the dashboard is also designed to hide data for skills with very low usage (e.g., under 1000 credits).
I'm logged in but receive a "Permission validation failed" error when I try to access my apps.
This error points to a temporary platform-side issue with validating user session permissions. It is typically resolved quickly by the engineering team.
Why are some of my credentials marked as "read-only"?
Credentials that are part of an organization-wide SSO integration are managed at a higher level and cannot be edited from within a specific workspace to maintain security. You can, however, create new personal or workspace-level credentials alongside them.
API & Integrations
This category focuses on issues related to using the DeepOpinion API and connecting to third-party services like Salesforce.
How can I generate an API Bearer Token if my company uses SSO?
Authenticating to the API requires a username and password, which is separate from the SSO login flow. Standard SSO credentials will not work. To use the API, you will need a dedicated non-SSO service account. Please contact support to request one.
I'm having trouble connecting to Salesforce with OAuth2.
If you have confirmed that your Salesforce "Connected App" settings are correct (e.g., by successfully connecting via Postman), the issue may be on the platform side. Please provide the URL of your automation workflow to the support team so they can investigate the specific integration setup.